RAGTruth

'RAGTruth' is a dataset/benchmark that contains checkable factual claims and is used to evaluate the faithfulness of generated responses in large language models by verifying these claims against provided evidence.

Papers using RAGTruth (9)

BALTO: Balanced Token-Level Policy Optimization for Hallucination Mitigation2026Evaluating the Relevance of Uncertainty Estimators for LLM Hallucination2026Hallucination Self-Play: Bootstrapping Reinforced Detector via Evolved Generator2026Detecting Contextual Hallucinations in LLMs with Frequency-Aware Attention2026Copy-Paste to Mitigate Large Language Model Hallucinations2025Turk-LettuceDetect: A Hallucination Detection Models for Turkish RAG Applications2025HalluGuard: Evidence-Grounded Small Reasoning Models to Mitigate Hallucinations in Retrieval-Augmented Generation2025Learning to Reason for Hallucination Span Detection2025100% Elimination of Hallucinations on RAGTruth for GPT-4 and GPT-3.5 Turbo2024